Design a Conversational AI Assistant

Design a large-scale conversational AI product similar to ChatGPT. Users should be able to: - start a new conversation, - send prompts, - receive streaming responses, - continue multi-turn conversations, - view chat history, - regenerate an answer, - optionally give feedback on responses. The system should support millions of users, low-latency response streaming, high availability, abuse prevention, and safe content handling. Focus on the online serving architecture rather than training a foundation model from scratch. Discuss: - functional and non-functional requirements, - APIs and core data model, - high-level architecture, - conversation memory and context-window management, - model routing and inference serving, - scaling, reliability, and cost control, - security, privacy, and moderation.